This achievement is due to three deep neural networks, artificial intelligence systems that mimic the brain
Observing their movements with the help of a camera, robots can learn the structure of their body and how they move and overcome problems they face, reveals research by Columbia University published in Nature Machine Intelligence.
In the study, the researchers have developed a way in which robots can model their own three -dimensional shapes by using a simple 2D normal camera. This achievement is due to three deep neural networks, artificial intelligence systems that mimic the brain. These concluded the 3D movement from the video allowing the robots to understand and adapt to their own movements, detect lesions in their bodies, and adjust their movements to recover from damage. As the researchers point out, as we deliver more and more critical functions to robots, from manufacturing to medical care, we need these robots to be more durable.
“As people learn to dance by watching their reflection in the mirror, so robots use raw video to obtain self -knowledge of their movements. Our goal is a robot who will understand his body, adapt to damage and learn new skills without constant planning from humans, “says Study’s lead author Yujang Hu, a doctoral student at Columbia University.
